Takaaki Fujita, OTR, PhD, Takuro Ohashi, OTR, Kazuhiro Yamane, OTR,
Yuichi Yamamoto, RPT, Toshimasa Sone, OTR, PhD, Yoko Ohira, MD, PhD,
Koji Otsuki, MD, PhD, Kazuaki Iokawa, OTR, PhD
Jpn J Compr Rehabil Sci 11: 28-34, 2020
Objective: To determine the lower limit of the number
of samples that is useful for creating a prediction
model on dressing independence in stroke patients by
using artificial neural networks.
Methods: Five datasets consisting of 120, 100, 80, 60,
and 40 were created from 121 stroke patients by
repeated random sampling. The models for predicting
independent dressing one month after admission were
created by an artificial neural network and logistic
regression in each dataset from the variables upon
admission to the convalescent rehabilitation ward. The
accuracy of both models was compared.
Results: The accuracy of the artificial neural network
model was significantly higher than that of the logistic
regression model in the 120, 100, and 80 patient
datasets, and there were no differences in the accuracy
of both models in the 60 and 40 patient datasets.
Conclusion: Our results suggested that the lower limit
of the number of samples for creating a useful
prediction model of dressing independence by using
artificial neural networks is approximately 80.
Key words: stroke, prediction, activities of daily living